A more exact formulation would be m^(1+o(1)) is equal to m^(1 + eps(m)) for some eps where eps(m) -> 0 for m -> infty
591 karma · joined June 2, 2016
A more exact formulation would be m^(1+o(1)) is equal to m^(1 + eps(m)) for some eps where eps(m) -> 0 for m -> infty
I made my partner switch to Latex thanks to the bibliography. Writing tables almost made her switch back to Word.
[0]: https://www.sfu.ca/~ljilja/cnl/info/UseCrossReference/index....
My partner, on the other hand, has a public contract (with teaching charge), she gets around 1 400€ after taxes. Living with this in Paris is not easy (1 studio of 15m² will cost around 650-700€)
The question is, how do you detect such a bias, and how do you prevent it in the first place? We have some ideas when it comes to humans, but for code it's much more challenging.
https://www.reuters.com/article/us-amazon-com-jobs-automatio...
The subject of AI and ethics is, in my opinion, something that is not considered enough by the actors of the field, including developers.
https://www.propublica.org/article/machine-bias-risk-assessm...
https://www.boredpanda.com/google-translate-sexist/
https://openai.com/blog/multimodal-neurons/#biasandovergener...
https://www.theverge.com/2018/1/12/16882408/google-racist-go...
In my opinion, e-voting is a solution to a problem that never existed.
Hence any cryptographic use of PCG is definitely prohibited
Take for instance N = 15, you can't find a log base 12 of 7, for instance (12 -> 9 -> 3 -> 6 -> 12). So you'd first need to find a fitting base, not sure how hard this is.
It has been proven [0] that factoring is as hard as solving DLP, but RSA also relies on the modular e-th root hypothesis, which we don't know to be as hard or easier than factoring.
In the end, because of [0] is you solve DLP you solve factoring yes, but we usually don't talk about DLP for RSA, even if as I said I was nitpicking.
[0] https://www2.eecs.berkeley.edu/Pubs/TechRpts/1984/5973.html
The reverse operation of RSA is the modular e-th root, not the discrete log.
Well, yes and no.
Yes, the bruteforce attack is a valid one. No, because usually the private key is around 256 bits, so you need to compute 2^255 hops on average to find the key.
The current biggest supercomputer can compute 10^17 flops/sec, assume you can do as many hops/sec. It still requires you 10^52 years to find the key.
The same is true for RSA : given a number N, find the two prime numbers p, q such that N = p * q. Trivial for N=15, a computer can do 13201225834171 = 1564571 * 8437601 in a matter of a second. However, current RSA N numbers have ~627 digits and we don't expect them to be factored in a reasonable amount of time.
As the other commenter said, there are much more efficient algorithms than bruteforce, but for current RSA keys (2048 bits) and ECC keys (256 bits), they are not reasonably efficient.
> Does this mean that a higher hop count is equal to a better private key?
So yes, the size of the key is a very important indicator of the strength of the key (but long keys can still be super weak if you don't generate them correctly). However bigger keys also result in longer encryption/decryption times.
A trapdoor function is when you have a function hard to invert (for any x, given y = f(x), find x), which inversion becomes very easy once you know some additional info. For instance in RSA, given c = m^e mod N, it is hard to find m. Unless you know d such that e*d = 1 mod phi(N), then you can easily find m by computing m = c^d mod N.
There is no known way of easily inverting exponentiation on finite groups.
To quote Wikipedia, "Functions related to the hardness of the discrete logarithm problem [...] are not known to be trapdoor functions, because there is no known "trapdoor" information about the group that enables the efficient computation of discrete logarithms. "
In Europe, you first study 3 years to get a bachelor's degree, then 2 more years for a master's degree (or an engineer degree), then you apply for a PhD which is supposed to last 3 years.
In natural sciences they do last 3 years in general, in human sciences they are usually quite longer.
There is no discussion whatsoever about this failure, except for "The low noise D-Wave 2000Q factored correctly all integers N up to L_N= 17"
The idea of the single scientist singlehandedly solving everything is mostly a myth. Most of the time, people rely on hundreds of previous papers before making their own contribution. Taken together, all the minimal steps make a significant leap.
Folding@home can very well be part of this effort.
I remember once, at Denfert-Rochereau, I had to let two RER go to the south without me, as there was litteraly no place for me to go inside. Could not even put a foot in.
Maybe this is to prevent adversarial learning? If the images reload immediately, then the bot can learn (via a neural network) whether its solution was good or not. If there's a delay, it's the same, but the learning is slowed down by the same factor.
No idea if this is true, it just popped out of my head.
From what I see this book is much more complete than a first year course, or even the whole curriculum of a classical CS education.
I'm quite familiar with math, but I never encountered wavelet theory, Gauss-Seidel method, Rayleigh-Ritz theorem, and many more. My knowledge about other subjects such as Hermitian spaces, quaternions, finite elements is quite superficial.
And I've only listed elements of Part I.
So indeed, we reach a contradiction, but the hypothesis was :
> To make things even easier, we can just throw out programs that loop infinitely or don’t return a 1 or a 0
so the contradiction tells us that \bar f is not in this set. More specifically, \bar f may loop infinitely at some point, I guess it does loop when you feed \bar f to itself.
* The DNN is actually getting close to the way human brain works, as it falls into the same mistakes as the human eye;
* There is some kind of universality in these illusions, that every (or most of) vision systems will fall for.
But honestly I think that the second conclusion is a rewriting of the first one (what is a vision system? What is this universal property?).
Again I'm absolutely not an expert of the field so please correct me if I'm wrong.
Also there may be a balance between "some hierarchy in the code so it can evolve" and adhoc code that just works for the current problem. Only having to write a new class for a new feature because everything else is handed smoothly is a bliss.
> According to the U.S. Geological Survey (USGS), the world’s volcanoes, both on land and undersea, generate about 200 million tons of carbon dioxide (CO2) annually, while our automotive and industrial activities cause some 24 billion tons of CO2 emissions every year worldwide
But I guess you're more knowledgeable on the subject than the US geological survey?
> On average, human activities put out in just three to five days, the equivalent amount of carbon dioxide that volcanoes produce globally each year
https://www.usgs.gov/news/human-activities-produce-more-carb...